Error calculation for a time series using autocorrelation

I'm trying to calculate the autocorrelation time for a time series generated from molecular-dynamics trajectory frames. I used several methods to estimate the autocorrelation time because I need it to compute the statistical error of an observable. My first attemp was block averaging. The block-averaging plot for the standard error of the mean (SEM) shows no clear plateau. The code I used for block-averaging:

sem=[]
            
            #return sems
            for i in range(1,1000):
                block_size.append(i)
                block_mean=[]
                no_blocks=0
                for l in range(0, len(nums),i):
                    if len(nums[l:l+i])==i:
                        block_mean.append(np.mean(nums[l:l+i]))
                        no_blocks=no_blocks+1
                sem.append(np.std(block_mean)/np.sqrt(no_blocks))
                

Example output (block-averaging SEM vs block size): My question: If there is no clear plateau in the SEM from block averaging, is it still meaningful to attempt to estimate the statistical error for these data? If so, what methods/precautions should I use to get a reliable error estimate (or a conservative bound) from a single long trajectory of Molecular Dynamics frames?

I tried using other methods such as statsmodel.tsa.stattools.acf for autocorrelation time calculation the error I end up getting is 0.0008712 which seems a bit unrealistic in my opinion. Is there a different way to calculate a more accurate error for these data?

I used the following code to calculate error using statsmodel.tsa.stattools.acf,

def tau_cal(acf_out,nums):
    positive_acf =np.where(acf_out > 0.0000)[0]
                
    result = [acf_out[x] for x in positive_acf]
    tau = 1 + 2 * np.sum(result[1:-1])
    Neff = len(nums)/tau
    error = (statistics.stdev(nums))/np.sqrt(Neff)
    return error
time = statsmodel.tsa.stattools.acf(nums)
error = tau_cal(acf_out,nums)